AI maturity model for agile delivery: Checklist with Excel template
Many AI maturity models are written for CIOs or enterprise programs. This level is often too high-level and too far removed from the realities of day-to-day agile software development for engineering managers.
This article translates existing AI maturity models into a pragmatic model for agile software development. In doing so, we are not chasing the AI hype, but following common sense along the thesis:
AI maturity in agile software development is reflected in whether AI accelerates and improves the value stream from problem understanding to user feedback.
Here you get 6 dimensions, each with 3 health check items for surveys and team retrospectives. At the end, you will also find an Excel template that summarizes all items as the basis for your maturity matrix.
TL;DR
- Classic AI maturity models tend to measure rather abstractly strategy, data, governance, engineering, operating model, culture, and value contribution.
- For agile software development, these dimensions should be translated into concrete capabilities of agile teams: goal clarity, knowledge context, verification, delivery system, collaboration, and continuous improvement.
- The best AI maturity measurement is not Excel reporting, but a basis for team retrospectives to implement concrete, tangible improvements.
What existing AI maturity models typically measure
AI maturity models usually follow a similar pattern: They measure an organization’s ability to embed AI strategically, securely, and effectively into work systems.
1. Technological capability levels
KPMG describes an AI capability model with six levels, from data readiness through prompting and contextualization to reliability, integration, and operationalization in scaled operations. This is technically useful because it makes clear that you do not need to worry about autonomous agents and scaled operations yet if there is still no robust data and context foundation.
Source: KPMG: AI Capability Maturity Assessment
2. Governance, risk, and lifecycle
OWASP AIMA is especially relevant for software teams because it views AI maturity along a lifecycle. The maturity model names strategy, design, implementation, operations, and governance as core domains. The linked Excel toolkit goes into even more detail and works with 8 practice areas: Responsible AI Principles, Governance, Data Management, Privacy, Design, Implementation, Verification, and Operations.
Sources: OWASP AI Maturity Assessment, OWASP AIMA PDF, OWASP AIMA Excel Toolkit
3. Pillars, heatmaps, and prioritized roadmaps
Gartner describes AI maturity as a diagnosis across several core pillars: strategy, data, governance, engineering, operating model, culture, as well as AI product and value contribution. The practical core is a heatmap between current and desired maturity, from which prioritized initiatives and roadmaps are derived.
Source: Gartner: AI Maturity Model and AI Roadmap Toolkit
4. Assessment, analysis, and joint workshop
AI Sweden describes maturity measurement as a three-step process: assessment, analysis, and workshop. The third stage is especially useful: stakeholders discuss the results together and develop a roadmap. This follow-up logic is exactly what many Excel templates lack (but of course not ours).
Source: AI Sweden: AI Maturity Assessment
For agile teams, this means: The retrospective is not an “add-on” after measuring maturity. The retrospective is the central place where insights are created from the maturity measurement and changes are initiated.
5. Ambition, capabilities, use cases, and implementation
Holisticon describes the appliedAI-related assessment across four dimensions: Ambitions, Capabilities & Enablers, Use Cases, and Execution. It is a good reminder that AI maturity is not just about governance. It takes ambition, capabilities, and relevant use cases.
Source: Holisticon: AI Maturity Assessment
Accenture also frames AI maturity as a transformation topic (and not just as tool adoption). Above all, this perspective confirms our article on AI transformation: AI maturity must be measured by the ability to change.
Source: Accenture: The Art of AI Maturity
What existing AI maturity models for agile software development are missing
Enterprise models are good for orientation. For engineering managers, however, they have three weaknesses:
- They are often too far removed from day-to-day team work.
- They measure many prerequisites, but too little delivery behavior.
- They easily create a maturity dashboard without improving the next team decision.
That is why I translate the models into 6 dimensions that a product or engineering team can actually discuss in retrospectives. Each dimension answers exactly one guiding question:
| Dimension | Guiding question | Function in the delivery system |
|---|---|---|
| Clarity of goals | Are we working on the right problems? | Direction |
| Shared knowledge context | Can AI understand our product and our domain? | Context |
| Verification & trust | Can we use AI results safely? | Trust |
| AI-adaptive delivery system | Does the team get better as a system? | Flow |
| Collaboration | Does AI become a team capability instead of individual optimization? | Alignment |
| Continuous Improvement & Governance | Do the organization and the rules get better with every iteration? | Learning loop |
The system behind it is simple:
- Goals determine what AI is used for.
- The knowledge context determines how well AI can work.
- Verification determines whether results are usable.
- The delivery system determines whether value is created faster as a result.
- Collaboration determines whether the team gets better together.
- Continuous improvement and governance determine whether improvements endure.
Logic of the AI maturity model: 6 dimensions and 3 levels for clear prioritization
For team retrospectives, I recommend 3 simple levels. Important: The levels do not first assess AI usage. They first assess the underlying delivery capability.
| Level | Meaning | Typical pattern |
|---|---|---|
| Level 1: Capability exists | The team basically masters the dimension. | Baseline |
| Level 2: Team practice established | The team has a shared practice for this dimension. | Repeatability |
| Level 3: AI integrated | AI systematically amplifies this capability. | AI-supported delivery impact |
How to use these levels: If Level 1 of a dimension is already problematic, first identify the problem with it and solve it. Once a healthy baseline has been established, you can move on to Level 2 and anchor the capability in the teams’ ways of working. Only when both Level 1 and 2 perform well does a focus on “AI integration” make sense. Of course, AI may already offer good solutions for Level 1 and 2, but AI should not yet be the mental focus there.
So here are the items for measuring the dimension, with the option to start the measurement directly with a retrospective in Echometer:
AI Maturity Assessment Template
Dimension 1: 🎯 Goal Clarity
This dimension checks whether AI improves work on the right problem. Many teams use AI for more output even though the problem, user need, or success criterion is vague. Then AI simply scales ambiguity.
AI maturity: 🎯 Goal clarity: How the retro works
Random Icebreaker (2-5 minutes)
Echometer provides you with a generator for random check-in questions.
Review of open actions (2-5 minutes)
Before starting with new topics, you should talk about what has become of the measures from past retrospectives to check their effectiveness. Echometer automatically lists all open action items from past retros.
Health Check
All team members can answer the health checks anonymously on a scale. Then go through the results of the health checks together and record any additional comments if necessary. If you use the same health checks in several retrospectives, you can also track trends over time in Echometer.
- Level 1: For our tasks, it is usually clear whether they have achieved their goal or not.
- Level 2: Before implementing topics, we always establish a shared understanding of the problem, solution, and success criterion.
- Level 3: AI systematically helps us understand user problems, weigh solution options, and define success criteria.
Discuss retro topics
Use the following open questions to collect your most important findings. First, everyone does it themselves, covered. Echometer allows you to reveal each column of the retro board individually in order to then present and group the feedback.
- What is currently holding us back in this dimension?
- What is the next best measure or next experiment to improve us in this dimension?
Catch-all question (Recommended)
So that other topics also have a place:
- What else would you like to talk about in the retro?
Prioritization / Voting (5 minutes)
On the retro board in Echometer, you can easily prioritize the feedback with voting. The voting is of course anonymous.
Define actions (10-20 minutes)
You can create a linked action via the plus symbol on a feedback. Not sure which measure would be the right one? Then open a whiteboard on the topic via the plus symbol instead to brainstorm root causes and possible measures.
Checkout / Closing (5 minutes)
Echometer enables you to collect anonymous feedback from the team on how helpful the retro was. This creates the ROTI score ("Return On Time Invested"), which you can track over time.
AI maturity: 🎯 Goal clarity
Health Check Questions (Scale)
Open questions
Good discussions often arise here around the question: “Which AI-accelerated work should we have never started in the first place?”
AI Maturity Assessment Template
Dimension 2: 🧠 Shared Knowledge Context
This dimension deliberately replaces the narrower term “data quality.” For agile delivery, it is not just about data, but about product knowledge, domain knowledge, architectural understanding, quality expectations, and shared decisions. AI can only do good work if this context is available and reliable.
AI maturity: 🧠 Shared knowledge context: How the retro works
Random Icebreaker (2-5 minutes)
Echometer provides you with a generator for random check-in questions.
Review of open actions (2-5 minutes)
Before starting with new topics, you should talk about what has become of the measures from past retrospectives to check their effectiveness. Echometer automatically lists all open action items from past retros.
Health Check
All team members can answer the health checks anonymously on a scale. Then go through the results of the health checks together and record any additional comments if necessary. If you use the same health checks in several retrospectives, you can also track trends over time in Echometer.
- Level 1: Relevant product and domain knowledge is readily available for my work.
- Level 2: As a team, we invest in a shared knowledge context that is up to date and usable for everyone.
- Level 3: AI helps us systematically identify knowledge gaps and ambiguities and improve context.
Discuss retro topics
Use the following open questions to collect your most important findings. First, everyone does it themselves, covered. Echometer allows you to reveal each column of the retro board individually in order to then present and group the feedback.
- What is currently holding us back in this dimension?
- What is the next best measure or next experiment to improve us in this dimension?
Catch-all question (Recommended)
So that other topics also have a place:
- What else would you like to talk about in the retro?
Prioritization / Voting (5 minutes)
On the retro board in Echometer, you can easily prioritize the feedback with voting. The voting is of course anonymous.
Define actions (10-20 minutes)
You can create a linked action via the plus symbol on a feedback. Not sure which measure would be the right one? Then open a whiteboard on the topic via the plus symbol instead to brainstorm root causes and possible measures.
Checkout / Closing (5 minutes)
Echometer enables you to collect anonymous feedback from the team on how helpful the retro was. This creates the ROTI score ("Return On Time Invested"), which you can track over time.
AI maturity: 🧠 Shared knowledge context
Health Check Questions (Scale)
Open questions
My opinion: For many teams, knowledge context is the underrated lever. Prompt training brings little value if team knowledge is scattered, outdated, or contradictory.
AI Maturity Assessment Template
Dimension 3: ✅ Verification & Trust
This dimension is at the core of AI maturity in software teams. AI can accelerate code, tests, acceptance criteria, analysis, and documentation. But only verifiable results may enter the value stream.
AI maturity: ✅ Verification & Trust: How the retro works
Random Icebreaker (2-5 minutes)
Echometer provides you with a generator for random check-in questions.
Review of open actions (2-5 minutes)
Before starting with new topics, you should talk about what has become of the measures from past retrospectives to check their effectiveness. Echometer automatically lists all open action items from past retros.
Health Check
All team members can answer the health checks anonymously on a scale. Then go through the results of the health checks together and record any additional comments if necessary. If you use the same health checks in several retrospectives, you can also track trends over time in Echometer.
- Level 1: I can reliably assess the quality of my work.
- Level 2: As a team, we have an established standard for good work that everyone adheres to.
- Level 3: With AI, we identify risks, errors, and quality gaps earlier and fix them faster.
Discuss retro topics
Use the following open questions to collect your most important findings. First, everyone does it themselves, covered. Echometer allows you to reveal each column of the retro board individually in order to then present and group the feedback.
- What is currently holding us back in this dimension?
- What is the next best measure or next experiment to improve us in this dimension?
Catch-all question (Recommended)
So that other topics also have a place:
- What else would you like to talk about in the retro?
Prioritization / Voting (5 minutes)
On the retro board in Echometer, you can easily prioritize the feedback with voting. The voting is of course anonymous.
Define actions (10-20 minutes)
You can create a linked action via the plus symbol on a feedback. Not sure which measure would be the right one? Then open a whiteboard on the topic via the plus symbol instead to brainstorm root causes and possible measures.
Checkout / Closing (5 minutes)
Echometer enables you to collect anonymous feedback from the team on how helpful the retro was. This creates the ROTI score ("Return On Time Invested"), which you can track over time.
AI maturity: ✅ Verification & Trust
Health Check Questions (Scale)
Open questions
A mature team does not ask: “Are we allowed to use AI for this?” It asks: “What evidence do we need in order to use this result responsibly?”
AI Maturity Assessment Template
Dimension 4: 🔁 AI-adaptive Delivery System
This dimension checks whether AI improves the value stream. Individual people may be faster while the overall system hardly improves at all. Then AI remains individual optimization. Maturity only emerges when the team adapts its way of working to the new possibilities.
AI maturity: 🔁 AI-adaptive delivery system: How the retro works
Random Icebreaker (2-5 minutes)
Echometer provides you with a generator for random check-in questions.
Review of open actions (2-5 minutes)
Before starting with new topics, you should talk about what has become of the measures from past retrospectives to check their effectiveness. Echometer automatically lists all open action items from past retros.
Health Check
All team members can answer the health checks anonymously on a scale. Then go through the results of the health checks together and record any additional comments if necessary. If you use the same health checks in several retrospectives, you can also track trends over time in Echometer.
- Level 1: Our team regularly delivers increments that are usable for customers.
- Level 2: Feedback loops with customers and the analysis of usage data are a fixed part of our team's value stream.
- Level 3: We actively use AI to turn usage data and user feedback into impact faster.
Discuss retro topics
Use the following open questions to collect your most important findings. First, everyone does it themselves, covered. Echometer allows you to reveal each column of the retro board individually in order to then present and group the feedback.
- What is currently holding us back in this dimension?
- What is the next best measure or next experiment to improve us in this dimension?
Catch-all question (Recommended)
So that other topics also have a place:
- What else would you like to talk about in the retro?
Prioritization / Voting (5 minutes)
On the retro board in Echometer, you can easily prioritize the feedback with voting. The voting is of course anonymous.
Define actions (10-20 minutes)
You can create a linked action via the plus symbol on a feedback. Not sure which measure would be the right one? Then open a whiteboard on the topic via the plus symbol instead to brainstorm root causes and possible measures.
Checkout / Closing (5 minutes)
Echometer enables you to collect anonymous feedback from the team on how helpful the retro was. This creates the ROTI score ("Return On Time Invested"), which you can track over time.
AI maturity: 🔁 AI-adaptive delivery system
Health Check Questions (Scale)
Open questions
The practical test: If AI disappeared from your work, would the value stream get worse or only the perceived productivity?
AI Maturity Assessment Template
Dimension 5: 🤝 Collaboration
This dimension is the blind spot of many AI maturity models. Agile software development lives on shared understanding, communication, decisions, and ownership. If AI is used only individually, it can even weaken teamwork: less shared context, less discussion, more parallel individual optimization.
AI maturity: 🤝 Collaboration: How the retro works
Random Icebreaker (2-5 minutes)
Echometer provides you with a generator for random check-in questions.
Review of open actions (2-5 minutes)
Before starting with new topics, you should talk about what has become of the measures from past retrospectives to check their effectiveness. Echometer automatically lists all open action items from past retros.
Health Check
All team members can answer the health checks anonymously on a scale. Then go through the results of the health checks together and record any additional comments if necessary. If you use the same health checks in several retrospectives, you can also track trends over time in Echometer.
- Level 1: I have a good overview of what is currently happening in the team.
- Level 2: Our communication within the team enables everyone to work effectively and stay up to date.
- Level 3: AI helps distribute relevant knowledge to the right people and reduces unnecessary information overhead.
Discuss retro topics
Use the following open questions to collect your most important findings. First, everyone does it themselves, covered. Echometer allows you to reveal each column of the retro board individually in order to then present and group the feedback.
- What is currently holding us back in this dimension?
- What is the next best measure or next experiment to improve us in this dimension?
Catch-all question (Recommended)
So that other topics also have a place:
- What else would you like to talk about in the retro?
Prioritization / Voting (5 minutes)
On the retro board in Echometer, you can easily prioritize the feedback with voting. The voting is of course anonymous.
Define actions (10-20 minutes)
You can create a linked action via the plus symbol on a feedback. Not sure which measure would be the right one? Then open a whiteboard on the topic via the plus symbol instead to brainstorm root causes and possible measures.
Checkout / Closing (5 minutes)
Echometer enables you to collect anonymous feedback from the team on how helpful the retro was. This creates the ROTI score ("Return On Time Invested"), which you can track over time.
AI maturity: 🤝 Collaboration
Health Check Questions (Scale)
Open questions
From my perspective, this is the most exciting difference compared with many enterprise models: an agile AI maturity model must measure whether AI makes the team better, not just individual specialists.
AI Maturity Assessment Template
Dimension 6: ☯️ Continuous Improvement & Governance
Governance is important, but it must not swallow everything. In this model, governance means: the team can make responsible decisions, make risks visible, and improve rules based on real experiences. Continuous improvement and governance belong together because rigid rules quickly become outdated in such a dynamic field.
AI maturity: ☯️ Continuous Improvement & Governance: How the retro works
Random Icebreaker (2-5 minutes)
Echometer provides you with a generator for random check-in questions.
Review of open actions (2-5 minutes)
Before starting with new topics, you should talk about what has become of the measures from past retrospectives to check their effectiveness. Echometer automatically lists all open action items from past retros.
Health Check
All team members can answer the health checks anonymously on a scale. Then go through the results of the health checks together and record any additional comments if necessary. If you use the same health checks in several retrospectives, you can also track trends over time in Echometer.
- Level 1: For my work, responsibilities and risk boundaries are clear at all times.
- Level 2: As a team, we regularly adapt the way we work based on new insights and lessons learned from experience.
- Level 3: AI helps us systematically question and further develop the way we work.
Discuss retro topics
Use the following open questions to collect your most important findings. First, everyone does it themselves, covered. Echometer allows you to reveal each column of the retro board individually in order to then present and group the feedback.
- What is currently holding us back in this dimension?
- What is the next best measure or next experiment to improve us in this dimension?
Catch-all question (Recommended)
So that other topics also have a place:
- What else would you like to talk about in the retro?
Prioritization / Voting (5 minutes)
On the retro board in Echometer, you can easily prioritize the feedback with voting. The voting is of course anonymous.
Define actions (10-20 minutes)
You can create a linked action via the plus symbol on a feedback. Not sure which measure would be the right one? Then open a whiteboard on the topic via the plus symbol instead to brainstorm root causes and possible measures.
Checkout / Closing (5 minutes)
Echometer enables you to collect anonymous feedback from the team on how helpful the retro was. This creates the ROTI score ("Return On Time Invested"), which you can track over time.
AI maturity: ☯️ Continuous Improvement & Governance
Health Check Questions (Scale)
Open questions
The goal is neither maximum freedom nor maximum control. The goal is a system in which teams can learn quickly without suppressing risks.
Tip: Simple AI maturity radar chart and heatmap with Echometer
Once you have covered all items with your team, you can prepare and visualize the data. Echometer even does that automatically for you:
If you conduct the AI maturity assessment for several teams, Echometer even provides a suitable AI maturity evaluation as a matrix / heatmap for the organization:
Therefore, my recommendation is: Instead of manual surveys and Excel, use Echometer so that you can benefit not only from professional analyses and trend analyses at the push of a button, but also from optimal support for facilitation and action tracking.
Excel template: All items of the AI maturity model for agile software development as a matrix
If you still want an Excel template for your AI maturity model, you can use the following Excel template:
| Dimension | Level | Survey item | Score 1-5 | Evidence | Biggest blocker | Next experiment | Owner | Review date |
|---|---|---|---|---|---|---|---|---|
| Clarity of goals | 1 | For our tasks, it is usually clear whether they have achieved their goal or not. | ||||||
| … |
Checklist: How to practically use the AI maturity model for agile software development
Don’t start with all 18 items in one huge assessment. Start with one dimension where you currently feel friction.
Each item is formulated as a simple agreement statement. If a team disagrees with Level 1, the basic capability is not yet stable. If Level 1 is true but Level 2 is not, a reliable team practice is missing. If Level 2 is true but Level 3 is not, AI is not yet a systematic amplifier of this capability.
So, here is your checklist for a smooth process:
- Choose one dimension that currently seems most relevant to you or your team. Focusing on all dimensions at once only leads to one thing: chaos.
- Have the team assess the three level items anonymously. For example, directly in Echometer’s retro tool.
- When evaluating, don’t discuss the average, but the deviations in your opinions. This reveals insights and makes opportunities visible.
- Also answer the two open questions in the retro template to develop a shared picture of blockers and possible measures.
- Formulate an experiment for 2 to 4 weeks. Agree on regular check-ins to ensure progress.
- After implementing the measure and an appropriate testing period, measure the same dimension again.
In addition to the checklist, a note on what you should definitely avoid is also allowed: If you compare multiple teams, compare patterns, not scores. A platform team, a product team, and a legacy team have different starting conditions. Maturity measurement becomes dangerous when it turns into a ranking.
More on this: Why agile maturity assessments often fail.
Conclusion: AI maturity is only useful if it also leads to improvements
Existing AI maturity models provide good building blocks: strategy, data, governance, engineering, operating model, culture, use cases, verification, and operations. For agile delivery, however, these building blocks must be translated into concrete capabilities of agile teams. That is why our items here offer practical and compact suggestions.
My recommendation: use Excel for overview, but use retrospectives for change. A team that honestly discusses one dimension and starts a good improvement (or even a good experiment) is further along than an organization with a perfect matrix and an extensive heatmap but no follow-through.
If you are looking for more input on AI in agile software development, these articles are a good next step:
FAQ on the AI maturity model for agile software development
What is an AI maturity model for agile software development?
An AI maturity model for agile software development evaluates how well a team translates AI into goal clarity, knowledge context, verification, the delivery system, collaboration, and continuous improvement. It measures not only tool usage, but whether AI improves the team’s value creation and learning ability.
How does it differ from classic AI maturity models?
Classic AI maturity models often look at enterprise perspectives such as strategy, data, governance, talent, operating model, and value contribution. For agile delivery, these dimensions must be translated into concrete capabilities of agile teams: better goal clarity, better knowledge context, reliable verification, an AI-adaptive delivery system, stronger collaboration, and continuous improvement.
Should I start with Excel or with a retrospective for an AI maturity assessment?
Start with a retrospective if you want to change behavior. Excel makes sense for documenting items, scores, evidence, and experiments. But the real insight comes from the conversation about blockers, risks, and the next small improvement step.
Why does the maturity model contain only three maturity levels?
Three levels are understandable and action-oriented for team retrospectives: capability present, team practice established, and AI integrated. A fourth level such as AI-native organization is useful as a vision in individual cases, but for many teams it is currently too distant to derive good, concrete measures.
